The future of mobile AI won’t be defined by algorithms alone, but by how well they navigate tightening privacy rules without losing speed or accuracy.
When I started working on in-app AI, I was fascinated by the huge amount of data insights it generated — patterns, behaviors, and insights that could transform how we understood user engagement.
We were building solutions for a market where app developers wanted to monetize through advertising but didn’t really know who their users actually were. Meanwhile, tech giants like Facebook, Google, and Twitter had a clear advantage — they could pinpoint audiences, serve hyper-relevant ads, and share profits with developers.
But then everything changed. Apple announced they would start removing the IDFA identifier — the unique ID that was consistent across the internet, allowing us to understand exactly who the user was. This wasn’t just a minor policy update; it fundamentally impacted our ability to group users into cohorts based on secondary criteria so ads could still be relevant without knowing the exact individual.
Behind closed doors, companies across the industry were scrambling to navigate the shift. We worked on solutions in real time, with no certainty about how the market — or users — would ultimately respond.
The reality of building under constraints
Working on these major platform changes taught me something crucial: the most sophisticated AI in the world means nothing if it can’t operate within real-world constraints.
We attended conferences — I remember traveling to Shanghai for one of the largest mobile conferences covering the Asian region — and everywhere, developers were asking the same questions about how to maintain personalization while respecting privacy.
We developed a range of product variations — from small-scale experiments that fine-tuned ad elements to boost clicks, installs, and overall performance, to more innovative solutions that hadn’t been tried quite that way before.
It wasn’t rocket science, but it was also something even major players hadn’t done. In fact, there are only a handful of companies offering these services, and at the time, we were all figuring it out together.
The key insight was that on-device AI wasn’t just a privacy compliance checkbox — it was actually a better way to build mobile experiences. When you process intelligence locally, you eliminate network latency, reduce server costs, and create applications that work regardless of connectivity.
Privacy, it turned out, wasn’t the end of innovation. It was the spark that made mobile experiences faster, cheaper, and more resilient.
From problem to solution: A personal journey
My understanding of these challenges deepened during COVID when everything shifted to remote work. I was spending about 60-70% of my working time on calls, coordinating with partners in different time zones, managing schedules that became incredibly tough. Prioritization became challenging — I couldn’t find time for workouts or other activities because of last-minute meetings.
This pain point led me to develop a productivity app focused on time management that would automatically identify free blocks in your schedule. But building an app that truly understands user behavior requires AI that operates seamlessly on-device. You can’t have an intelligent calendar assistant that needs to phone home every time someone opens their calendar.
The technical reality is that mobile devices have strict memory and battery constraints, but they also have increasingly powerful neural processing capabilities. The challenge isn’t whether you can run AI on mobile — it’s whether you can do it efficiently enough that users actually want to keep using your app.
Technical lessons from the field
During my time prototyping new products and doing extensive research, several key principles emerged for on-device AI optimization:
Start with score-based models: Rather than trying to run full neural networks on every interaction, develop models that generate relevance scores locally. These scores can inform decision-making without exposing sensitive user data or requiring constant server communication.
Design for interruption: Mobile users don’t interact with apps continuously. Your AI needs to maintain context across sporadic usage patterns while being efficient enough to activate quickly when needed.
Test across real conditions: We conducted thorough UX research to see how people responded, examining how users reacted to different levels of information and what kind of flow could be established. The laboratory environment tells you nothing about how AI performs when someone’s battery is at 15% and they’re on a spotty cellular connection.
The broader industry transformation
What I witnessed working with major ad platforms wasn’t just a privacy update — it was a fundamental shift in how the mobile industry approaches user intelligence.
The companies that will succeed are those that can deliver personalized experiences while operating within privacy constraints and mobile resource limitations.
This transformation is particularly critical for emerging markets and regions with limited connectivity. During my experience, I realized that solutions developed for high-bandwidth, always-connected environments often fail when deployed globally. On-device AI naturally addresses these limitations.
The market opportunity is enormous. We worked on products that were outside the typical industry focus, mostly because traditional advertising companies don’t approach mobile AI this way. There’s space for innovation, but it requires understanding both the technical constraints and the user experience implications.
The key insight from my experience is that the future of mobile AI isn’t about porting desktop or server AI to phones — it’s about rethinking intelligence for mobile-native experiences. This means designing AI that enhances rather than interrupts user workflows, respects privacy by default, and operates efficiently within mobile constraints.
For developers entering this space, my advice is simple: understand your constraints first, then build intelligence that works within them. The most elegant AI architecture means nothing if it drains your user’s battery or requires permissions they’re uncomfortable granting.
The privacy wars in mobile advertising taught us that user trust is the ultimate constraint. Build AI that earns and maintains that trust, and you’ll create experiences that users actually want to engage with long-term.
___
About the author
Nikolay Sidiropulo is internationally recognized for his contributions to mobile AI optimization and user experience applications, blending years of hands-on work with major advertising platforms and cutting-edge mobile app development.
